diff --git a/python/test/test_api.py b/python/test/test_api.py
index 0f4a22a94..a6adde27d 100644
--- a/python/test/test_api.py
+++ b/python/test/test_api.py
@@ -345,11 +345,6 @@ def test_ready(self):
         server = tritonserver.Server(self._server_options).start()
         self.assertTrue(server.ready())
 
-    @pytest.mark.xfail(
-        tritonserver.__version__ <= "2.46.0",
-        reason="Known issue on stop: Exit timeout expired. Exiting immediately",
-        raises=tritonserver.InternalError,
-    )
     def test_stop(self):
         server = tritonserver.Server(self._server_options).start(wait_until_ready=True)
 
@@ -455,15 +450,20 @@ def test_basic_inference(self):
             "bool_input": numpy.random.rand(1, 100).astype(dtype=numpy.bool_),
         }
 
-        for response in server.model("test").infer(
+        response_iterator = server.model("test").infer(
             inputs=inputs,
             output_memory_type="cpu",
             raise_on_error=True,
-        ):
-            for input_name, input_value in inputs.items():
-                output_value = response.outputs[input_name.replace("input", "output")]
-                output_value = numpy.from_dlpack(output_value)
-                numpy.testing.assert_array_equal(input_value, output_value)
+        )
+
+        responses = list(response_iterator)
+        self.assertTrue(len(responses), 1)
+        response = responses[0]
+
+        for input_name, input_value in inputs.items():
+            output_value = response.outputs[input_name.replace("input", "output")]
+            output_value = numpy.from_dlpack(output_value)
+            numpy.testing.assert_array_equal(input_value, output_value)
 
         # test normal bool
         inputs = {"bool_input": [[True, False, False, True]]}
diff --git a/python/tritonserver/_api/_response.py b/python/tritonserver/_api/_response.py
index 8dc505298..60418dc64 100644
--- a/python/tritonserver/_api/_response.py
+++ b/python/tritonserver/_api/_response.py
@@ -29,7 +29,6 @@
 from __future__ import annotations
 
 import asyncio
-import inspect
 import queue
 from dataclasses import asdict, dataclass, field
 from typing import TYPE_CHECKING, Optional
@@ -182,9 +181,6 @@ def _response_callback(self, response, flags, unused):
                 asyncio.run_coroutine_threadsafe(
                     self._user_queue.put(response), self._loop
                 )
-            if flags == TRITONSERVER_ResponseCompleteFlag.FINAL:
-                del self._request
-                self._request = None
         except Exception as e:
             message = f"Catastrophic failure in response callback: {e}"
             LogMessage(LogLevel.ERROR, message)
@@ -308,9 +304,6 @@ def _response_callback(self, response, flags, unused):
             self._queue.put(response)
             if self._user_queue is not None:
                 self._user_queue.put(response)
-            if flags == TRITONSERVER_ResponseCompleteFlag.FINAL:
-                del self._request
-                self._request = None
         except Exception as e:
             message = f"Catastrophic failure in response callback: {e}"
             LogMessage(LogLevel.ERROR, message)